Visualizing ontologies on the web

Transcription

1 Visualizing ontologies on the web Ioannis Papadakis 1, and Michalis Stefanidakis 2 1 Ionian University, Department of Archives and Library Sciences, Plateia Eleytherias, Corfu Greece 2 Ionian University, Department of Computer Science, Plateia Eleytherias, Corfu Greece {papadakis,mistral}@ionio.gr Abstract. The proposed paper introduces an ontology visualization model suitable for average web users. It is capable of retaining ontology expressiveness while at the same time hiding the formal terminology that is usually employed in the context of ontology development. According to the proposed ontology visualization model, classes are represented as boxes and their corresponding properties are represented as labeled lines connecting such boxes. By hoping from one box to another, users are able to interactively explore the underlying ontology. Such interaction is facilitated through an intuitive, web-based GUI implemented in Javascript, which communicates with the underlying ontology through a middleware component implemented in Python. Keywords: semantic web, owl, ontologies, visualization 1 Introduction Ontologies nowadays constitute the backbone of what is wider known as the semantic web, defined in 2001 by Tim Berners Lee [1]. However, ontologies have been around for quite some time, long before they found their way to the semantic web. Researchers belonging to the Knowledge Engineers KE community, employed ontologies to conceptualize various domains in order to be able to reason with such domains through asserted and/or inferred knowledge. In this context, various applications have emerged capable of providing the necessary infrastructure for developing ontologies. Such applications are basically standalone software components capable of visualizing ontologies mostly for editing purposes. Since the advent of the semantic web, an ever increasing number of researchers not necessarily belonging to the KE community became interested in ontologies. They could be addressed as information experts in various domains wishing to formally model their expertise in the context of a semantic web application. Despite the fact that semantic web applications are regrettably difficult to find on the web, a significant number of ontologies has been produced. However, current on-

2 2 tology visualization systems are mostly addressed to KE experts rather than average web users. The employed vocabulary is only familiar to the KE community and visualization techniques rarely make use of most recent web technologies. The proposed paper aims in providing an ontology visualization model suitable for average web users. It is argued to be capable of retaining ontology expressiveness while at the same time hiding the formal terminology that is usually employed in the context of ontology development. The underlying ontology is visualized as a linear list of nodes that are being interactively chosen by users. The rest of this paper is structured as follows: The next section presents a number of existing efforts in visualizing ontologies. The proposed ontology visualization model is introduced at the next section. In order to test the model in action, two case studies have been conducted. The two case studies are presented in the two following sections respectively. The first one visualizes an ontology for the library domain while the second one visualizes an ontology for the academic domain. Each ontology has its own special features highlighting various features of the ontology visualization model. Finally, this paper draws conclusions and points directions for future work concerning the proposed ontology visualization model. 2 Related work in ontology visualization In order to bring the semantic web closer to the public, the underlying information structures should be visualized in a way that is easily comprehensible from average users. In this context, the problem of visualizing ontologies should be approached from a Human-Computer Interaction (HCI) perspective. The vast majority of ontology visualization tools are based on combinations of form-based GUIs with hierarchical trees (e.g protégé, ontoxpl [4], ontoweb [5]) or on graphs with classes as nodes and arcs as slots (e.g. jambalaya, ontoviz [6]) or as clusters (Cluster Map [7], SWHi [2]) or, more recently, on crop circles [8]. Such visualization styles have proved to be suitable for KE experts, wishing to engineer ontologies. However, it seems that it is very difficult for average users on the web to take advantage of the expressiveness of ontologies through the aforementioned tools. KE specific vocabulary finds its way to the screen, resulting in causing frustration to non-experts visitors. Most frequently, after a while, users are overwhelmed with numerous widgets that are squashed in their screen, flooding this way users short-term memory, thus infringing basic HCI principles [9]. Moreover, the aforementioned solutions are delivered mostly in client-side applications that cannot be easily integrated to the overall web infrastructure. Consequently, dissemination is limited to closed communities instead of the wider web population. It is the authors' belief that nowadays, wide dissemination of ontology visualization applications can only be achieved through common web technologies such as the web server and web browser. Indeed, users are already accustomed to the web and consequently to the corresponding metaphors that have been established.

3 It should be kept in mind though, that designing an elegant, functional user interface on the web is not an easy task. So far, conventional sites have to deal with the so-called 'page-reload effect', which refers to the fact that whenever a user triggers a request to the server, the whole page has to reload on screen regardless of the amount of new information that arrives from the server, as compared to the previous state of the site. Such problems drove web designers to build light-weight sites, in order to minimize the page-reload effect. Another problem that web designers have to deal with, refers to the fact that not all browsers support the same versions of de facto web standards such as (X)HTML and Javascript. Thus, in order to remain interoperable and compliant with the majority of web browsers, web designers have to be extremely cautious with the applications they develop. Moreover, many technologies promised at their beginning to revolutionize the web. However, history shows that such technologies eventually failed to fulfill their promises due to various reasons. For example, Java-applets emerged as a way to integrate desktop applications within the web browser environment. Despite the initial success, Java-applets failed due to browser incompatibility, slow download and start up of applets, unpredictable behavior on different operating systems, lack of browser support and no standard security model 1. Current trends on the web indicate that many of the above issues can be solved through the employment of Ajax [3] technology. Such approach promotes user interaction by utilizing asynchronous http requests to the server, thus eliminating the page-reload effect. Moreover, Ajax applications rely heavily on Javascript, which, after a rather long period of existence, finally seems to be well supported from major web browser vendors. Having the above thoughts in mind, the ontology visualization model that is presented in this paper is based on common web technologies. As it will be described in the following sections, the GUI that delivers the proposed functionality to average web users, interactively visualizes the underlying ontology through the employment of common HTML widgets and the most promising Ajax technology. 3 3 Proposed ontology visualization model The proposed ontology visualization model is comprised of boxes that correspond to classes, and labelled lines corresponding to properties linking such classes. Users may hop from one class to a related one by selecting one of the various properties each class is being explicitly and/or implicitly (through a reasoner) involved. Classes related to a given class through a user-selected property are grouped together within a pop-up context menu. Such a menu is created on demand by clicking on a property. Properties are outlined inside each box. Finally, by selecting an item (i.e. class) within a context menu, a new box is drawn to- 1 Available at:

4 4 gether with a line that connects the two boxes. Individuals require applicationspecific elaboration. The whole iterative process is illustrated in Fig. 1. Class A Class C Class D property 1 property 11 property 6 property 2 property 3 property 4 property 3 property 3 property 7 property 8 property 22 property 7 property 3 property 2 Class A Class F Class B Fig. 1. Ontology visualization model According to wikipedia 2, in the context of Computer Science (and Information science), ontologies are data models representing concepts and their corresponding relations within a specific domain. Such a definition constitutes ontologies as highly expressive information schemas. However, as argued earlier in this paper, it seems that ontology visualization research is more concerned in finding ways of aiding researchers building ontologies than finding ways to expose the underlying knowledge to average users. An average web user does not care whether a specific ontology classification derives from asserted or inferred statements. Additionally, it is equally irrelevant for an average user the type of a property and/or a restriction (e.g. symmetric, functional, etc). All it matters is not why but what kind of relations exist for a given concept (formalized as a class). In this context, the proposed visualization model is addressed to average web users instead of KE experts. Thus, the employed vocabulary doesn t contain any formal terminology. Classes and properties are represented as boxes and labeled lines respectively, complying this way with well-known metaphors at the web. The simplicity of the approach does not imply that the employment of more sophisticated tools such as reasoners and/or query engines (e.g. sparql) is not possible. The following sections illustrate the effectiveness of the proposed visualization model through the presentation of two case studies, referring to a simple and a more complex ontology respectively. 2 Available at: accessed at:

5 4 Case study 1: Visualizing an ontology of Library s Subject Headings According to this case study, average web users are able to explore an ontology that is comprised of subject headings existing within the Online Public Access Catalog s OPAC s library of the Computer Science Department at the Ionian University in Greece 3. Libraries are very important memory organizations containing information assets that are most of the times formally described. The descriptive information of each information asset participating in the library s OPAC is of premium quality, since it is produced by experienced staff (i.e. librarians). Perhaps the most important semantic descriptor within a library s catalog is the subject descriptor. In order to register an information asset within the library s OPAC, among other things, librarians have to attach one or more suitable subject headings to the particular asset. Such headings are usually constructed from wellknown subject headings registries such as the Library of Congress Subject Headings LCSHs (available at: authorities.loc.gov/). According to current practice from OPAC vendors, users have access to such information through a form based interface where users are prompted to select subject from a drop-down menu and accordingly provide a value to the adjacent textbox that best describes their conceptual information needs (see fig. 2a). After that, the OPAC performs a Boolean search to a subject-based index in order to provide users with information assets displayed as a typical search results list. Upon selection of a particular information asset, users are able to observe the subject headings of the particular information asset and, accordingly, click on the one that best matches their information needs. This way, a search for information assets indexed under the specified subject heading is performed (see fig. 2b). 5 Fig. 2a. Performing subject-based search: step one 3 Available at:

6 6 Fig. 2b. Performing subject-based search: step two The aforementioned process may be repeated until users find the information assets they are looking for. Such a scenario indicates that there is no way of exploring the subject headings according to their underlying structure, without having to visit specific information assets. Thus, finding information assets about a particular subject becomes a rather tedious process, despite the availability of the required knowledge. The application presented in this case study aims at simplifying such a process. In order to test the proposed ontology visualization model, an ontology capable of modeling the underlying subject headings was created. The ontology contains a total of about 500 LCSH s records (terms), corresponding to the subject headings employed within the Ionian University s library. Such terms constitute the official subject headings (and synonyms) defined by the Library of Congress or their Greek translations as translated from trained librarians within the library. The ontology is encoded in OWL-DL format and models subject headings as classes, and relations as object properties. There are four relations defined, namely contains, is_part_of, incontextof and seealso. Each relation corresponds to a property. Such properties act as restrictions to specific named classes, according to the official LCSHs. The ontology treats the label of a subject heading as well as all of its synonyms and Greek translations as individuals. The ontology is accessed from an application server implemented in python. The application server receives queries in XML format from the client-side and delivers responses in XML format. The client-side is implemented in javascript based on the XMLHttpRequest XHR [3] object. Users interact with the clientside and transparently address queries to the application server. 4.1 User interaction Upon initialization of the ontology visualization model, users are presented with a subject heading visualized as a box consisting of a title (i.e. formal LCSH heading in English), possible subtitles (i.e. alternative subject heading and/or translated subject heading in Greek) and one or more entries corresponding to the named relations of the formal subject heading. Users are able to right-click on a relation in order to observe in a pop-up context menu the related subject headings (see fig. 3a). By clicking on a subject heading within the context menu, a new box is drawn at the right, containing in-

7 formation about the selected subject heading. The two boxes are connected with a labeled line representing the relation (see fig. 3b). Simultaneously, subject headings of the most right box appear in the OPAC s search box. Thus, a search query is addressed to the underlying search engine, which returns a list of information assets being described by the selected subject heading (see fig. 3c). The proposed semantic web application dictates that individuals exist both as alternative headings inside each box and as query phrases within the underlying search engine s search box. 7 Fig. 3a. Linear list ontology visualization Fig. 3b. User interaction Fig. 3c. Application - specific individual elaboration The proposed visualization model as applied to the specific case study for the library domain, aids average web users in locating information assets about a given subject heading. Users decide which part of the underlying ontology will be presented to them by selecting the appropriate class. It is argued that the employment of common web widgets (i.e. labeled line, box, context menu) enables average users to transparently explore the underlying on-

8 8 tology. Users are unaware of the ontology per se, while at the same time they are able to take advantage of it s expressiveness through the employment of visualization widgets. Moreover, despite the rather large size of the ontology, the screen is never overcrowded with too much information. Despite the fact that user interactions may lead to the construction of numerous boxes, user s interest is always at the right side of the screen, containing the most recent class metaphor. Another interesting feature of the described application is the fact that it promotes serendipitous or casual discovery of interesting information (see [10]). During the ontology browsing process, users underpin their cognitive learning, since they are able to witness which information assets correspond to the formally defined subject headings they visit. 5 Case study 2: Visualizing an ontology referring to a University s Department According to this case study 4, average web users are able to explore an ontology that refers to the Computer Science Department at the University of Piraeus in Greece. As compared to the previous case study, this ontology is more expressive. It consists of nineteen classes organized in three levels of abstraction, six object properties, one datatype property and seventy nine individuals. Object properties, although not organized into an hierarchy, may have attributes and/or relations (e. g. propa inverse of propb). The proposed visualization model presents each class as a box containing relations that correspond to the properties the individuals of this class participate in. Users may click on a relation in order to come up with a list of its corresponding classes and accordingly decide which class will be visualized at the right of the existing box. It should be noticed that according to the proposed model, a relation between two classes may correspond to either an asserted or inferred view of the class hierarchy. Users do not need to know the origin of the class hierarchy (i.e. asserted or inferred). In order to accomplish such a task, the ontology is classified offline by a reasoner and stored within a suitable data structure. The internal structure acts as an index and contains information about each ontology asset (i.e. class, property, individual) deriving both from asserted and inferred knowledge. This way, the resulting semantic web application achieves tolerable response times during user interaction. Otherwise, real time invocation of a reasoner would result in considerable delays during the interactive ontology visualization process, discouraging this way user interaction. As far as elaboration of individuals is concerned, the proposed case study follows the search results list paradigm (see fig. 4). 4 Available at:

9 9 Fig. 4. Ontology visualization model as applied to case study 2 Specifically, in the case of hierarchical relations (i.e. contains, ispartof), individuals belonging to the user-selected class are grouped together in a list beneath the linear list of boxes. Similarly, in the case of pair of classes that are connected through the property a user has selected, the search results list contains the corresponding pairs of individuals. Again, the type of a relation (i.e. attribute of a property) is irrelevant to the average user. 6 Conclusions This paper defines an ontology visualization model suitable for average web users. The proposed model intends to bring ontologies closer to the web community. It is the authors belief that ontologies are highly expressive information structures capable of being visualized in a way that they remain expressive while at the same time they hide their complexity from average web users. Ongoing research in ontology management tools, results in applications that are targeted towards KE experts wishing to develop ontologies. Such applications provide ontology visualization models that contain domain-specific terminology, and employ visualization widgets that are most of the times unfamiliar to average web users. The proposed model breaks down ontologies to their fundamental ingredients, namely classes, properties and individuals. Such components are visualized through common widgets that are well-established on the web. This way, average users are able to transparently explore ontologies. Ontologies are represented as dynamic lists of classes interconnected with their corresponding properties. Users are able to interact with such lists deciding this way which part of an ontology will be presented to them. Thus, users are always in control of the ontology visualization process. The effectiveness of the proposed model is demonstrated through the presentation of two case studies relying on two fundamentally different ontologies respectively. Both case studies refer to web-based applications targeted towards average users.

10 10 Future work includes the application of the proposed ontology visualization model to a number of existing ontologies, discovering this way possible limitations. Moreover, the authors would like to extend the expressiveness of the model by exposing a bigger part of the underlying ontology to the web-based GUI. Benchmarking the effectiveness of the model in terms of web usage (i.e. scalability, response time, usability) is also under way. Finally, research is being conducted in finding ways of visualizing individuals in a more consistent manner. 7 Acknowledgements The authors would like to acknowledge the work of Katerina Tzali and Kyriaki Papoulia on helping developing the ontologies of the first and second case study respectively. References 1. Berners-Lee, T., Hendler J., and Lassila O.: The Semantic Web. Scientific American 284, no. 5, 34-38, (May 2001) 2. Fahmi, I.: SWHi System Description: A Case Study in Information Retrieval, Inference, and Visualization in the Semantic Web. Proceedings of European Semantic Web Conference, Austria, (2007) 3. Garett, J.J.: Adaptive Path Ajax: A New Approach to Web Applications, Adaptive Path Essay Archive, available at: Haarslev, V., Lu, Y., Shiri, N.: OntoXpl: Exploration of OWL Ontologies. Proceedings of the Web Intelligence Conference WI 2004, Beijing, IEEE Computer Society (2004), ISBN Giunchiglia, F., Gomez-Perez, A., Stuckenschmidt, H., Pease, A., Sure, Y. and Willmott, S.: Ontology-based information exchange for knowledge management and electronic commerce. Proceedings of 1 st International Workshop on Ontologies and Distributed Systems (ODS2003), Mexico, (August 2003) 6. Sintek, M.: The Ontoviz Tab: Visualizing Protégé Ontologies. available at: 7. Fluit F., Sabou C., van Harmelen F.: Ontology-based information visualization: Towards semantic web applications. Springer Verlag, (2005) 8. Wang, T.D., Parsia, B.: Cropcircles: topology sensitive visualization of owl class hierarchies, T. D. Wang and B. Parsia, Proceedings of International Semantic Web Conference (ISWC 06), (2006) 9. Bates, M.J. Where Should the Person Stop and the Information Search Interface Start? Information Processing & Management 26, , (1990). 10. García, E., Sicilia, M.A. Designing Ontology-Based Interactive Information Retrieval Systems. Proc. of the Workshop on Human Computer Interface for Semantic Web and Web Applications, Springer LNCS 2889, , (2003)